Problem in optimization with reaxFF potential

[ramkrishna]

Hi,
When I am running a structure using reaxFF potential, it produces "nan" in the output whereas it is not showing any error when I am using  pedone potential for the same structure.
Is it not possible to use reaxFF potential to execute structure optimization or MD ? or am I doing anything wrong in the input script? Please find the attached input and log files. Your kind help would be greatly appreciated.
Thanks
Ramkrishna





 

[Ulrik G. Vej-Hansen]

Seems strange to get nan's - I will look into it.

[ramkrishna]

Dear Ulrik,
I would really appreciate if you get a chance to look into this.

Waiting for your valuable reply!

Thanks
Ramkrishna

[Ulrik G. Vej-Hansen]

Sorry about the delay, I need to run it by the relevant developers. I will get back to you as soon as possible.

[Anders Blom]

The selected ReaxFF potential does not work for this system; it's missing certain bonds that are necessary. If you try it in the "Quick optimizer" in the Builder you do get a proper error message about this, and also if you use the potential in a script as

potentialSet = ReaxFF_CHONSSiPtZrNiCuCo_2005()
calculator = TremoloXCalculator(parameters=potentialSet)

Your modification to set "strict_bondpairs = False" means you bypass the warning, but the NaN is precisely why the warning is there in the first place.

Btw, let me also note that the system as constructed doesn't make much sense to optimize... If you really want a Cu/ZrO interface you need a lot more ZrO on one side and Cu on the other.

[ramkrishna]

Hello Anders,
Thank you so much for your reply. Is there any way to modify this potential to include all those missing bonds?
Actually, I wanted to put this system (smaller size) as an example structure. If it works, then of course I will go for longer system. 

Thanks
Ramkrishna

[Anders Blom]

Short answer is no. You could of course extend the potential, by fitting the interaction parameters of those bond pairs to DFT, but that's huge work and requires specific expertise on ReaxFF (which not even we have internally).